presentation real-time adaptive radiotherapy...

BETTER CANCER CARE FOR OUR COMMUNITY Jeremy.Booth@health.nsw.gov.au AAPM • San Antonio • 18 July 2019

Real-time adaptive radiotherapy

Jeremy BoothNorthern Sydney Cancer Centre, Royal North Shore Hospital,

Sydney Australia

Institute of Medical Physics, University of Sydney

Overview

• Definitions/context

• What’s in the box?

• How well does it work?

• Future?

Real-time adaptive definition

• To adapt radiotherapy delivery to account for intra-treatment anatomic change - during treatment– May include inter-fraction variation implicitly

Obtain real-time position/pose

Predict future position/pose

Adapt treatment (beam, couch)

Some more definitions

• Real-time: the latency or response time of motion is less than the timescale of significant changes in the position during the response time

• Latency: time lag between target motion and the complete execution of the system response

• Temporal prediction: estimation of target position at some time in the future ~latency

• Tumour trailing: delivery adapts to baseline shifts

• Tracking: increasingly used to refer to monitoring motion, has/is used to refer to adaptation

Clinical drivers for real-time adaptation

• Safety

Patient-connected

Deliver planned dose

• Efficient workflow / Automation

Improve consistency

• Compact dose / Accuracy

Reduce OAR dose

MLC TRACKINGDYNAMIC BEAM

Lung SABR dose accuracy: IndividualPLANNED MLC TRACKING

PTV = GTV +5 mm

PLANNED ITV-BASEDPTV = ITV +5 mm

DELIVERED MLC TRACKING DELIVERED ITV-BASED

PTV - MLC Tracking

PTV - ITV-based

PTV = red contour; dose cloud >95% dose

Relative dose (%)

Lung tracking reduces PTV/MLD

• Guides towards safety for retreatment, larger tumours, central tumours, higher target doses

System Peak-to-peak motion (cm)

PTV reduction

MLD reduction

Reference

Cyberknife 0.2 – 2.3 16 - 49% Nr Chan et al 2013

Vero 0.2 – 2.6 16 - 53% Nr Depuydt et al 2014

MLC tracking 0.3 -1.8 2 - 47% 7 - 38% Booth et al ESTRO 2019

MLC tracking meets the challenge of dose painting

Red = PTV

Colvill IJROBP 2015

MLC tracking for dose painting moving targets (prostate, liver, pancreas)

High dose off GTV

Target dose not delivered

High dose GTV defined from mMRI + PSMA PET

Isotoxic Liver SABR prescription

• Liver SABR – Goal: 50Gy/5#

– Target dose constrained by normal liver tolerance

• Gargett etal simulated gating/tracking– 13/20 cases had constrained target dose

– 11/13 cases reached target dose with real-time adaptation

– Further dose escalation possible

StandardFree-breathing 42.5Gy/5#

Motion management 50Gy/5#

Gargett et al RadOnc 2019

Real-time adaptive action loop

Real-time tumour monitoring systems

Calypso &Micropos

Electromagnetic

Implanted fiducial markers

Radioisotope

Navotek

Optical surface markers

VisionRT

kV/MV/ Optical

and a combination of them … Slide courtesy of Paul Keall

Volumetric

MRIlinac

Ultrasound

Overview of Technologies

• Motion monitoring (tracking) method• Type of information

• Point, shape, volume• Surrogate, target/OAR

• Frequency of information

• Adaptation method– Gating, Robotics, MLC tracking, Couch tracking, Gimbal tracking,

Combinations– Generally incorporates the prediction component

History of real-time adaptation

2019201020001990 1995

CK Spine tracking 2005

Couch tracking, feasibility 2005

MLC tracking lung 2016

Gated treatments 1989

CK Lung tracking 2002, First patient 2004

CK Markerlesslung tracking 2007

RadiXact2019

MLC tracking prostate 2014

Research

Unity MLC tracking 2019

Commercial milestones

MLC tracking concept 2001

MLC tracking linac/phantom 2007

Gated treatments

1999 (RT-RT)

Vero 2014

Gating ~2004

Gating

• Gating stops the treatment when the target moves outside a pre-defined tolerance eg 2mm

• Available clinically with a range of motion triggers– First papers from 1989 with research systems

• Abdominal compression sensors, RT-RT

• Commonly applied to SABR techniques, DIBH

• Duty cycle is the proportion of time beam in on and is a compromise to residual motion

• Reference: TG76 (2006), currently in revision

RT-RT system

Varian RPM

Robotics: AccuRay Cyberknife

• Utilises dual kV and optical

• Robotic arm sychronises linactargeting with detected motion

• Motion model correlates optical surface with target motion, updates model regularly

• Synchrony system detects direct tumour location (markerless)

• Planning on single phase of 4DCT

Couch Tracking

• Couch position adapts to synchronise with motion– Single target

– 6DoF

– Range of couch motion

• Not commercialised, no patients treated

Gimbal tracking (VERO)

• Detects motion using dual kV and optical surface

• Tracks with gimbal ring gantry and couch

• Utilises a motion model for correlation of external to internal motion

• No longer in production

MLC tracking

• MLC adapts shape to synchronise with motion– Real-time optimisation of leaf position

• Area of under dose and over dose

– Unique ability to track deformation and multiple targets

– Conformity linked to leaf speed, width

– Accuracy dependent on plan complexity

• Planning on single phase of 4DCT, no ITV

Deformation

Lung and nodeProstate and node (ethics filed)Image-based

tracks DVFGe et al 2014

Example MLC tracking control loop

Clinical status of MLC tracking

Prostate Oct 2013 – June 2015Calypso-guidance

▫ 28 patients (858 fractions)▫ V7.4 MLC controller

Lung SABR July 2015 – Oct 2018Calypso-guidance

▫ 17 patients (70 fractions)▫ V7.5 MLC controller▫ Prediction on

Prostate SABR Nov ‘16 – Jul ‘18KIM-guidance

▫ 10 patients (50 fractions)

All with University of Sydney MLC tracking software (latency 220ms); All with Varian Trilogy linac, Millennium MLC

MR-guided MLC tracking

• MRI volumetric tracking

• Superb soft tissue contrast

• An MLC tracking algorithm developed at Royal Marsden/DKFZ for deformable targets (Fast et al 2014)

• Implemented on Elekta Unity (Glitzner et al 2019)– Latency of 347ms/4Hz and 204/8Hz.

• Potential for real-time functional imaging

Combination - RadiXact

• Radixact detects motion from kV and surface

• Utilising Synchrony to track lung, liver, pancreas…

• Treats with helical beam

• RadiXact adapts with jaw, binary MLC and couch

• Latency 70ms Schnarr et al 2018

Uncertainty for real-time adaptive RT

• 4DCT & lack of 4D planning• Tumour delineation• Surrogate vs direct tumour tracking• Latency vs Prediction accuracy• MLC tracking accuracy

– Real-time monitoring accuracy – Leaf fitting (leaf width)– Leaf adjustment (leaf speed)

Understanding MLC tracking

• Target localization error– caused by system latency and

uncertainties in the monitoring system

• Leaf fitting error– occurs when the desired MLC shape is

fitted onto the closest physically possible MLC shape and is caused by the finite leaf width.

• Leaf adjustment error– the difference between the actual and

the requested leaf positions and is caused by the finite leaf speed. Poulsen et al IJROBP 2012

Clinical Challenges from 4DCT

• Planned on exhale phase of 4DCT– 4D treatment with 3D planning

• 4DCT can under estimate motion– Leads to potential geographic miss

• 4DCT can over estimate motion– Leads to larger PTV, higher lung dose

• ?4D planning

• ?Improved 4DCT image quality

MLC tracking lung Patient 2: beacon motion in SI direction during treatment

Clinical challenges with fiducials

1. Potential toxicity with implantation– Bronchoscopic under II guidance: 0/17 toxicity

– Benchmarking markerless tracking (Shieh et al 2017)

2. Fiducials provide a surrogate for tumour position– Surrogacy at 4DCT and treatment (fluoro)

– Mean = 2mm, max=4mm

– 4DCT motion not always representative

Surrogacy from fluoro (AP)

Contour GTV on all phases

Find beacons on each phase

Surrogacy from 4DCT

Slide courtesy of Nick Hardcastle

QA of real-time adaptive system

• Published guides for Cyberknife/Synchrony and MLC tracking

• System integrity– End to end test with hidden target and known motion(s)

– Latency measurement

• Real-time QA– Interlocks for unexpected conditions ie target outside

tracking area, comms error, large/fast motion

– Potential for EPID-based aperture check or back-projected dose estimation

• Retrospective dose reconstruction

Future Outlook

• Emerging technologies– Faster data processing / algorithms

– Artificial intelligence• Automation

• Bridge gaps in information

– Evolution of geometric margins towards probabilistic

– Real-time dose prediction dose-guided adaptation

– Real-time functional imaging biology-guided adaptation

Summary

• Real-time adaptation available for 30 years (gating) and 20 years (tracking)– currently first generation: adapting to rigid motion of target

• Real-time adaptation well aligned to emerging technologies– Automation, machine learning, real-time volumetric imaging

– Implicitly includes ‘plan of the day’ / inter-fraction change

• Next generations will consider:– Deforming anatomy (tumour and OAR)

– Dose/biology-guidance

presentation real-time adaptive radiotherapy...

Documents

z z á á á v u t s w æ u w r x - ohio auditor of...

Æ ] u v ( } x d z x í z w z Ç ] > } } Ç 6lqjoh 6olw ......

2020 rules and regulations working (1) - …d>^ } ( ' ] v ]...

z z}µo } zu u]v] } z}µo u]v] x - lou holtz

Æ ] u v ( } x d z x í z w z Ç ] > } } Ç

&>h/ e > dzk>zd > e r r î · 2020. 4. 12. · u ] v v v v...

(x ,y ,z ) (u ,v ) (x ,y ,z ) (u ,v ) (x ,y ,z )

^ Ç v z ] v ] } Ç v z ] ^ µ ] } v Ç } u ] ] v - thieme^...

í x ^d z Á î t & z } µ z ] d } v k&&/ / > zh> ^ î x...

ri /hqw 0dufk · ô w ï ì x u x : } z v u z o d z } } &...

terranova grade 5 practice test - floassets.blob.core...

nr. x z w x v w z

7jtb #bmbodf 5sbotgfs€¦ · í x ì ì 9 } ( z ( } ] p v...

d } k ] u ] ] ( } z e Æ í î d } v z expectation...

^ Á ] z } v } o · Á ] z x ^ Ç ] v p ] v z ] u ( ] v v...

} µ µ ^ d } o w ] } } µ µ e u dzw ~z x zw ~z x '^d ~z...

v v Ç & ] v v ( } / v o µ ] À ' } Á z w ] À > ] u ] ~z...

v À ] ] sk...v v d z ] u v µ odh^d p ] À v } z µ } ( z }...

v o ] ] } v w ì ô x ì í x î ì í õ / v ] ^ µ } ( > ]...

pdf of keynote · 2020-06-25 · v ( } ] À > z ] x o o z ]...